With plans to archive it for 100 years, they can just simply wait until vulnerabilities in block ciphers are discovered (possibly by the NSA) and hold on to the data until then and decrypt it.
I'm sure they're decrypting everything they know how, but for everything else, it's just archival.
Remember, it was the NSA that designed most of them to begin with and still hires the best of the best to continue the work. It's also probably a lot easier to just tap into everything and record it for later decryption if necessary than try to target various networks or systems individually.
It might be possible that the NSA has discovered some side-channel attack [2]. If that's what you mean by backdoor, then I suppose you could be right. (Perhaps every CPU has been manufactured so as to facilitate timing attacks.) I still think that the probabilities combine to some astronomically low number because those attacks still take time, and I don't think they have the computational ability to simultaneously run attacks against any very large subset of the US population.
I don't trust secret, unaccountable organizations... but it just seems too extraordinary to imagine that any organization could actively monitor billions (millions, even) of communication streams at a time.
[0] http://en.wikipedia.org/wiki/AES_implementations
[1] http://csrc.nist.gov/archive/aes/rijndael/Rijndael-ammended....
Maybe this is just how masterfully deceitful they are.
I also don't understand the cost-benefit ratio associated with actively spying on every transaction in the US. The vast majority of adversaries are outside the US, clustered in a handful of countries. I know that the US ran massive domestic anti-communism operations during the Cold War, but that would have been far cheaper than setting up a data center capable of capturing, storing, and processing every electronic transaction. (And even those Cold War operations didn't focus on every American.)
It reminds me of clients who think I can predict basically anything if I have enough data and computational power. Most of them seem to think that I can automate the process, that they don't need to be involved, that it doesn't take weeks of gut feeling analysis, and that it'll basically give them God-like omniscience.
I spend a lot of time educating customers about the limits of "big data". If the federal government is running such a program, they need to hire a consultant for education's sake.
Edit: Also, reading the post about NSL's [2], I get the impression that the NSA is still at the same level of tech as the general public. Why wouldn't they just use Stellar Wind?
[1] http://www.wired.com/threatlevel/2012/03/ff_nsadatacenter/
[2] https://www.noisebridge.net/pipermail/noisebridge-discuss/20...